Research On Fault Diagnosis Method Of Rolling Bearing Based On Variational Modal Decomposition

According to statistics,the equipment failure caused by rolling bearing fault accounts for a large proportion in the total failure.Therefore,it is necessary to carry out regular inspection or real-time monitoring of the bearing state to avoid the damage of other components caused by rolling bearing failure.Although the intelligent diagnosis method supported by big data avoids the disassembly of mechanical equipment,the fault recognition rate needs to be improved.Therefore,it is urgent to optimize the diagnosis method to improve the fault recognition rate.First of all,after exploring the influence of different modal numbers and penalty factors on the variational mode decomposition,this paper proposes a method to determine the modal number through the JS divergence value of the reconstructed signal and the original signal spectrum.After the mode number is determined,an optimization function of the penalty factor is determined according to the fluctuation degree of the central frequency of each component.When the two parameters are optimized and applied to the actual signal decomposition,the problem caused by improper parameter setting can be effectively avoided.Moreover,after the fault signal is decomposed,the time domain diagram of component appears more obvious impact.Secondly,according to kurtosis criterion,the components after VMD decomposition are reconstructed,and the theoretical characteristics of time domain,frequency domain and entropy of the reconstructed signal are obtained.In order to avoid the redundancy of features and the low discrimination of the selected features to the four fault states,three types of features of the reconstructed signal are screened by the Laplace score,and the two features with the minimum Laplacian score are formed into the feature vector.Then,after analyzing the principle of SVM algorithm,we know that the parameters and will affect the classification accuracy of SVM.Therefore,bat algorithm is used to optimize the two parameters of SVM,and the classification performance of ba-svm is tested with two kinds of data in UCI data set.The results show that ba-svm can achieve higher classification accuracy than SVM model with random parameters.Finally,the feature vectors extracted from parameter optimized VMD,parameter determined VMD and EMD are input into BA-SVM for training.The classification results of test samples show that when the optimized VMD is applied to three kinds of data sets,the samples under four fault states are basically not misclassified,so the recognition accuracy of fault state is much higher than that of parameter determined VMD and EMD.